Investigate and remove `Rf_findVarInFrame3` usage in `environment`

[DavisVaughan]

https://github.com/r-lib/cpp11/blob/1c9dbb65cc9279a404f035f2fca990b338c9f602/inst/include/cpp11/environment.hpp#L37-L51

[DavisVaughan]

Likely going to use R_existsVarInFrame() in exists() on R >=4.2.0

---

And then likely going to use the new R_getVarEx() on >=R-devel, although it has different semantics than Rf_findVarInFrame3(env, sym, TRUE) in edge cases, i.e.:

• For R_MissingArg values in the env, R_getVarEx() errors
• For promises in the env, R_getVarEx() will evaluate them before returning them

I don't think there is anything we can do about this unless we get an iterator API for environments that doesn't add special behavior like this

[DavisVaughan]

Lionel and I decided that for the proxy SEXP conversion method, we should use the new R_getVar() function (not the extended version), which would:

• Error on R_MissingArg (definitely what R Core wants from us)
• Error on unbound symbols in the env (you should check with exists() first, also good to not expose R_UnboundValue)
• Evaluate promises (probably what the typical user wants)

https://github.com/wch/r-source/blob/0986e80068addc5c884d06580412ff2331436814/src/main/envir.c#L2305

And then for the fallback for older R versions we call Rf_findVarInFrame3() but then add 3 checks to reimplement the 3 bullets from above, so the behavior is consistent on all R versions.

After the enum / iterator API for environments is released, people will have more options if they want to manually implement more advanced behaviors than this "safe" way.